Live intelligence

Latest AI updates

1 updates
Research

LLM agent redesigns magnesium stent across ~1,650 CAD parameters within 15 simulations

An arXiv preprint reports that a general-purpose LLM (referred to as GPT-6 Astra) edited a magnesium stent through a CAD parametrisation of roughly 1,650 values while limited to a budget of 15 nonlinear finite-element simulations. Within that budget it lowered the objective, a high percentile of strains in the stent's bends, by 32%, whereas the industrial shape optimiser Tosca improved the original design by about a tenth and then stopped. An earlier LLM design, 25% lower on the same objective, was validated in an independent higher-fidelity full-deployment model with crimping and damage, showing 14% less strain than the original and 63% less damage at full expansion. The authors argue LLMs can combine learned knowledge with simulation feedback to coordinate large-scale redesign rather than local surface polishing.

LLM agents for designCAD parameter optimizationnonlinear finite-element simulation